A novel method for detection of abnormal conditions during plant operation uses wauelet-domain hidden Markov models (HMMs) as a powerful tool for statistical modeling of wavelet coefficients. By capturing the interdependence of wavelet coefficients of a measured process variable, a classification strategy is developed that can detect abnormal conditions and classify the process behavior on-line. The method is extended to include multiple measured variables in detection and classification. Two case studies illustrate the potential of this method.